On linear convergence of gradient-type minimization algorithms